The Acquisition of a Phonological System

As yet, the present chapter focused on infants’ speech sound discrimination skills. In the preceding sections, it became clear that early perception de-velops from a universal perception of speech contrasts to a language-specific one during the first year of life. Although infants’ perceptual and discrimina-tive capacities are not yet adult-like at the end of this period (e.g., Kuijpers 1996; Polkaet al. 2001), 12-month-old infants seem to perceive speech con-trasts mostly in line with the phoneme categories of their mother tongue.

As adumbrated already in section 4.2, infants’ contrast discrimination skills in the first year of life are rather of phonetic than of truly phonological nature. Although infants’ perception has adapted to the native phoneme inventory at around 10 to 12 months of age, phoneme contrasts must not only be perceivable but they also have to be stored in long-term memory and

4.5 The Acquisition of a Phonological System 95 they must be available in distinguishing minimal pairs in order to success-fully establish and process meaning distinctions. Since the present thesis is mainly concerned with infants’ pure discrimination skills and not with their phonological proficiency, the topic of phonological acquisition will not be discussed as elaborately as the preceding sections on very early speech perception. The current section will provide a rough sketch of linguistic de-velopment in the second year of life and in particular of infants’ distinction of phoneme contrasts.

At the centre of the following discussion is the finding that infants’ ca-pacity to phonetically discriminate phonemic contrasts does not necessarily entail successful application of the same contrast in more complex tasks at the beginning of the second year of life. Stager & Werker (1997) use the Switch Procedure in a series of experiments with Canadian English infants to investigate whether they are able to detect a subtle phonemic contrast – the PoA contrast in the syllables ‘bih’ [bI] and ‘dih’ [dI] – and whether the infants are able to use this contrast in word learning. The results reveal that 14-month-olds do not recognise the native /b/–/d/ contrast in a word-learning task, neither if the infants are familiarised with two word-object pairings in the habituation phase (Experiment 1), nor when task demands are reduced by presenting only one word-object combination during habit-uation (Experiment 2). Infants aged 8 months, however, have no problems in discriminating ‘bih’ and ‘dih’ (tested in the variant with one word-object pairing only). Two additional experiments attest that 14-month-olds are generally capable of accomplishing the task and that they did not com-pletely lose the ability to discriminate the PoA contrast in question. The infants detect a ‘switch’ between two phonetically dissimilar non-words (‘lif’

[lIf] and ‘neem’ [ni:m]) in a single word-object association task. Further-more, 14-month-olds succeed in distinguishing ‘bih’ from ‘dih’ when the visual stimulus presented during habituation and test trials is a checker-board pattern instead of one of the fancy objects presented in the previous experiments.²⁷ Stager & Werker (1997) explain the discrepancies in the contrast detection skills between 8- and 14-month-olds by suggesting that the task is different for the two age groups, respectively. The 8-month-olds, still in a ‘pre-phonological’ phase of language acquisition, pass the experi-ment as a plain sound discrimination task with the object on the screen

27It is assumed that due to its static, screen-filling nature and its plain, two-dimensional appearance a checkerboard pattern is unlikely to be taken as an object. Accordingly, infants are assumed not to associate the pattern with a label. The checkerboard image is meant to be only a visual attractor for infants’ gazes, thus making the Switch task a pure syllable discrimination task (see appendix A for more details on the Switch Procedure).

serving as a simple attention-getter.²⁸ The older infants are more advanced in their linguistic skills and have begun to understand the referential nature of language. Accordingly, they attempt to establish meaningful connections between auditory and visual stimuli, that is, for them the experiment is a word-learning task. According to Stager & Werker (1997), linking words and objects is a difficult task for 14-month-old infants, which is demanding more cognitive resources than a simple discrimination task. Consequently, they assume that children who have just begun to build up a mental lexicon are very busy concentrating on the lexical aspects when engaged in a word learning task, hence they neglect fine phonetic detail. In the syllable pair

‘bih’–‘dih’ phonetic differences are minimal and seem to be disregarded. The phonetic contrast between the non-words ‘lif’ and ‘neem’ is larger, hence the 14-month-olds’ success in the word-learning task.

The findings by Stager & Werker (1997) show that although Canadian English 14-month-olds are capable of distinguishing the native PoA con-trast between /b/ and /d/ in a discrimination paradigm, they fail to use fine phonetic detail in a word-learning task, that is, when they are required to establish meaningful links between words and objects. A study by Pa-ter et al. (2004) shows that 14-month-olds’ inability to distinguish native phoneme contrasts in word-learning tasks is not restricted to the ‘bih’–‘dih’

contrast. In order to make the stimuli conform to English phonotactics, the authors add a coda to the test syllables, making the stimuli ‘bin’[bIn]and

‘din’ [dIn]. The addition of a final consonant has no effect on infants’ per-formance in the word-learning task: 14-month-olds fail to notice a ‘switch’

between ‘bin’ and ‘din’ in a Switch experiment just as they fail to distin-guish ‘bih’ and ‘dih’ in Stager & Werker’s (1997) study. Moreover, Pater et al. (2004) find that 14-month-old infants fail to distinguish a laryngeal contrast between ‘bin’ [bIn] and ‘pin’ [p^hIn] as well as a combination of a PoA and a laryngeal contrast (‘din’–‘pin’) in a simplified variant of the Switch Procedure with only one word-object pairing.²⁹ The replication and expansion of Stager & Werker’s (1997) results suggest that 14-month-old English-learning children do not only fail to distinguish the PoA contrast

28As a side note it shall be emphasised that speaking of a ‘pre-phonological’ phase is intended to refer to a phase before infants start linking words and objects and building a lexicon. ‘Phonological’ in this context is not meant to say something about the existence of phonological representations. As will be seen below, it is sometimes suggested that infants have detailed phonological representations even though they have not yet entered the word-learning stage (see section 4.5.3).

29Phonetically, the laryngeal contrasts are described as “(partially) voiced unaspirated”

versus “voiceless aspirated” (Pateret al.2004:392,394).

4.5 The Acquisition of a Phonological System 97 between /b/ and /d/, but that the differentiation of fine phonemic contrasts in general poses a problem for 14-month-olds when they are engaged in a word-learning task. Apparently, early phonological systems – if one wants to speak of a ‘phonological system’ already – are not completely reliable yet.

This discovery, which will also be referred to as the ‘bin/din-paradox’

below, leads to a debate about the nature of early phonological representa-tions which is far from being settled and which cannot be traced in detail here. Making no claim to be complete, some exemplary positions will be picked out and presented in the following paragraphs. A central question arising from the findings by Stager & Werker (1997) and Pateret al.(2004) is whether infants fail to distinguish phoneme contrasts in word-learning tasks because they lack the phonological fundament or because they are somehow hampered in using or accessing their phonological knowledge. In other words, is it a problem of competence or performance?

4.5.1 The Resource Limitation Hypothesis

By relating infants’ phoneme distinction capacities to processing load, Stager

& Werker (1997; see also Pateret al.2004) are proponents of the performance argument. Under their view, thediscrimination ability in the checkerboard-variant of the Switch Procedure demonstrates that the phoneme contrast is in place. Yet, since the construction of label-object associations is computa-tionally more demanding than simple speech sound discrimination, infants fail to use their knowledge in word-learning experiments, in which their pro-cessing capacities are drawn upon otherwise. Such a view, also referred to as resource limitation hypothesis (Fennell & Werker 2003), entails the as-sumption of continuity between perceptual or phonetic representations and lexical/phonological representations. Following the parsimony principle, the idea that an already existing (phonetic) representation has to be built anew as a phonological one when infants start to process speech signals lexically is rejected. If information is discriminable phonetically it must also be avail-able for phonological use. If, however, the phonetically discriminavail-able infor-mation cannot be utilised for phonological processing, this is assumed to be due to specific performance limitations and not to a discontinuity between phonetic and phonological development.

Different phenomena observed in a series of follow-up studies provide support for theresource limitation hypothesis. First of all, infants’ inability to distinguish phoneme contrasts in a word-learning situation seems to be confined to only a short period in children’s development. Besides repli-cating Stager & Werker’s (1997) finding of English-learning 14-month-olds’

failure to distinguish ‘bih’ from ‘dih’ in a word-learning task with the Switch Procedure, a study by Werkeret al.(2002) shows that 17- and 20-month-old Canadian English toddlers have no difficulties detecting the ‘switch’. Fur-thermore, they find that 14- and 17-month-olds’ performance in the word-learning task correlates positively with their vocabulary size. More advanced word learners are better at distinguishing the phoneme contrast in the per-ception experiment. From these findings the authors conclude that there is only a brief temporal window when infants are unable to distinguish fine phonetic detail in word-learning tasks. When infants are just at the begin-ning of learbegin-ning to link words and objects, the task demands of the Switch Procedure are computationally too complex, leading to a neglect of fine phonetic detail. When children become more used to word-learning, access to phonetic detail is no longer limited and thus the 17- and 20-month-old participants succeed in distinguishing ‘bih’ and ‘dih’.

Furthermore, it has been shown that a reduction of the task demands can result in successful learning and subsequent perceptual distinction of pho-netically similar words. Also using the Switch Procedure, Fennell & Werker (2003) test Canadian English 14-month-olds’ ability to distinguish the same PoA contrast as employed in Stager & Werker’s (1997) study, that is, /b/

versus /d/. In contrast to the previous studies, the stimuli consist of famil-iar words and objects (/b/all vs. /d/oll with according pictures as visual stimuli)³⁰. In this case, the infants have no difficulties in distinguishing /d/

from /b/; they reliably notice the ‘switch’. The authors explain this finding by reference to the differential tasks. While the studies by Stager & Werker (1997), Werker et al. (2002) and Pateret al. (2004) require infants to learn new (non-)words to detect the ‘switch’, infants in Fennell & Werker’s (2003) study are already familiar with the words presented in the test. Thus, in the latter study, infants are no longer tested on their word-learning skills but rather on their word-recognition skills. Consequently, they do not have to focus their attention on the linking of objects and labels. Instead, the processing load is reduced, leaving more cognitive resources to accessing phonetic detail.

These assumptions are confirmed by various experiments using the Pref-erential Looking Paradigm (PLP).³¹ Swingley & Aslin (2002) demonstrate

30Fennell & Werker (2003) test infants learning Canadian English as spoken in Vancou-ver. According to the authors, in this variety of English, the words ‘ball’ and ‘doll’ form a minimal pair and can be transcribed as[bO:l]and[dO:l].

31From a methodological point of view, the PLP is principally less demanding than the Switch Procedure. While the latter method requires infants to keep the objects and their labels in memory, the PLP offers to choose between two different pictures. Even if the

4.5 The Acquisition of a Phonological System 99 that 14- to 15-month-old American English infants are sensitive to phonetic detail in a word-recognition task. Infants notice mispronunciations in conso-nants and vowels of familiar words, suggesting that their representations are fairly detailed and that they use this detail to distinguish words. Similarly, Ballem & Plunkett (2005) find that 14-month-old infants learning British English are more sensitive to segmental mispronunciations in familiar words (e.g., [p^hOl]for[bOl] ‘ball’) than in newly learned non-words (e.g., [duk] for [t^huk] ‘tuke’). In line with the findings by Werker and her colleagues and by Swingley & Aslin (2002), they suggest that the representations of newly learned words are still rather fragile and prone to be affected by high task demands whereas the representations of familiar words have become more robust already. Older infants with more linguistic experience are assumed to be more advanced in their speech processing skills so that word learning is less of a cognitive challenge to them, leading to successful phoneme distinc-tion in word-learning tasks, an assumpdistinc-tion that is in line with the findings of Werkeret al. (2002; also cf. Swingley & Aslin 2000; Swingley 2003 and references there).

Another way of reducing task demands is to make infants more famil-iar with the objects whose labels are to be learned. In a study by Fennell

& Werker (2004) 14-month-old Canadian English infants without explicit knowledge of the word ‘doll’ are given some time in advance of the virtual experiment to play with the doll later shown on the screen. In the Switch Procedure (with the picture of the doll being the sole visual stimulus), they succeed in distinguishing ‘doll’ from ‘goll’. A facilitation of the task is also achieved by emphasising the referential connection between the visual and the auditory stimuli, that is, by making the learning situation more explicit.

Yoshida et al. (2009) test 14-month-old infants from an English language background with a mixture of the Switch Procedure and the PLP (calling it a ‘visual choice paradigm’). The learning phase is the usual habituation pro-cedure known from the Switch experiments. Infants are familiarised with the syllables ‘bin’ and ‘din’, each presented in combination with a fancy object as visual stimulus. In the test trials, both objects appear on the screen, but only one of them is named. If infants learn the labels and are able to distin-guish the minimal pair they will look longer to the correct object. Moreover, filler items of known words and objects are presented between the test tri-als, referred to in a carrier sentence making the connection between visual and auditory stimuli more explicit (e.g., “Car . . . Look at the car!”; Yoshida

child may not be absolutely certain, one of the pictures will be a better match to the label heard in the test trial than the competitor image (see, e.g., Yoshidaet al.2009).

et al.2009:414). The emphasis of the referential nature of the task as well as the forced choice testing situation apparently lead to a reduction of process-ing load. Accordprocess-ingly, and in contrast to the findprocess-ings by Stager & Werker (1997), Werker et al. (2002) and Pater et al. (2004), Yoshida et al. (2009) report that 14-month-olds succeed in learning the similar-sounding syllable pair ‘bin’–‘din’. Similarly, Fennell & Waxman (2010) find that 14-month-olds from an American English language background succeed in learning and distinguishing ‘bin’ and ‘din’ when the referential audio-video-stimuli relation is made more explicit than in the classical Switch Procedure.

Finally, a study by Thiessen (2007) suggests that word learning in a Switch experiment is also eased when the phoneme contrast to be tested appears in a lexically dissimilar context in the familiarisation phase (Exper-iment 2). During habituation, American English-learning 15-month-olds are presented with the three synthesised (non-)words ‘daw’, ‘tawgoo’ and ‘daw-bow’, each paired with a fancy object as visual stimulus. In the test phase, infants succeed in distinguishing ‘daw’ [tO:] (ACT: 5 ms) and ‘taw’ [t^hO:]

(ACT: 47 ms), both paired with the ‘daw-object’ known from the habitua-tion phase. However, if the contrast appears in a similar lexical context dur-ing habituation (Experiment 3: ‘daw’, ‘tawgoo’, ‘dawgoo’), 15-month-olds fail to distinguish ‘daw’ from ‘taw’ in the test phase. The study by Thiessen (2007) suggests that it is not the existence of minimal pairs or, more gen-erally, lexical density in the mental lexicon that forces infants to establish detailed phonological specifications. Instead, contrast learning seems to be more difficult if the phonemes in question occur in similar contexts.

So far, the studies presented in the current section provide support for the assumption that at the age of 14 months, infants in principle have the phonological basis to distinguish minimal pairs, to learn new words and to store them in memory. Their inability to access the relevant information seems to be due to a temporary processing overload caused by specific task demands. As soon as the experimental requirements are decreased and also when infants become older and more experienced with language learning, they no longer have difficulties in learning and distinguishing minimal pairs.

What is calledresource limitation hypothesis by Werker and her colleagues can be seen as a description of a phenomenon that is otherwise labeled

‘computational bottleneck’ and which is not restricted to phonology and word learning. Such performance limitations seem to occur also in other areas of the language acquisition process and in child development in general, as Crain & Lillo-Martin (2003:27) suggest:

4.5 The Acquisition of a Phonological System 101

“These bottlenecks can be attributed to some limitation in a non-linguistic cognitive capacity, such as attention span or memory span.

The idea of a computational bottleneck also influences adult behavior in some circumstances. It is difficult to do two demanding tasks simul-taneously – such as driving a car down a twisty road and discussing a philosophical issue. Apparently, children have such bottlenecks at a much lower level.”

The assumption that infants’ failure in phoneme distinction is due to perfor-mance limitations and not a problem of competence implies that phoneme representations are in place at the age of 14 months. The studies mentioned above assume that infants’ language-specific perception at the end of the first year of life provides evidence for the existence of native phoneme cat-egories, but they do not explicitly elaborate on how these representations come into existence and how exactly they look like. The issue will be ad-dressed below, but prior to that, some approaches arguing in favour of a competence explanation shall provide a counter-perspective to theresource limitation hypothesis.

4.5.2 Fragmentary Phonological Systems

Competence explanations are based on the assumption of discontinuity be-tween perceptual phonetic representations on the one hand and phonologi-cal representations on the other hand. The two types of representations are considered to belong to two distinct systems which are acquired separately.

For instance, Brown & Matthews (1997; as summarised in Fennell & Werker 2004) suggest that phonetic and phonological developments are independent of each other and that they even move in opposite directions. The phonetic repertoire is supposed to be rather broad and universal at the beginning of infants’ life. Gradually, infants stop paying attention to those contrasts that do not occur systematically in the sound signal they are exposed to.

This development is described as ‘pruning’.³² In contrast to the decremen-tal pattern of phonetic development, the phonological system and with it lexical representations are assumed to develop in the opposite direction, as

This development is described as ‘pruning’.³² In contrast to the decremen-tal pattern of phonetic development, the phonological system and with it lexical representations are assumed to develop in the opposite direction, as